On using temporal features to create more accurate human-activity classifiers

Through advances in sensing technology, a huge amount of data is available to context-aware applications. A major challenge is extracting features of this data that correlate to high-level human activities. Time, while being semantically rich and an essentially free source of information, has not received sufficient attention for this task. In this paper, we examine the potential for taking temporal features—inherent in human activities—into account when classifying them. Preliminary experiments using the PlaceLab dataset show that absolute time and temporal relationships between activities can improve the accuracy of activity classifiers.